Socially assistive robots (SARs) are designed to support people through social interaction without providing direct physical assistance [1,2]. In health care, they have been explored as guides, companions, educators, telepresence platforms, and communication mediators [3-5]. Hospitals are especially relevant settings because they combine logistical complexity with emotional strain, repeated information needs, and heavy staff workload [3,6,7].

Pediatric outpatient environments add a further layer of complexity. Children often face uncertainty, waiting, unfamiliar spaces, and procedure-related stress, while parents must simultaneously manage navigation, appointments, and the child’s emotional state. Prior reviews suggest that social robots can support children’s well-being during medical treatment through distraction, engagement, and emotionally lighter interaction, but the evidence remains heterogeneous and strongly context dependent [8,9]. Hospital human-robot interaction (HRI) reviews likewise show that pediatric settings are promising, while many studies remain short-term, small-scale, or conducted under controlled conditions instead of routine clinical conditions [3,7].

Implementation is an additional challenge. Health care robot adoption is rarely determined by technical capability alone. Acceptance depends on whether users understand what the robot is for; whether it fits work processes; whether it is easy to use; and whether organizations provide the training, maintenance, and responsibility structures needed for sustained use [10-12]. Recent European survey evidence also suggests that public support for SARs in health care is conditional. People appear more comfortable with bounded assistive roles, such as information provision and guidance, than with higher-stakes gatekeeping or quasi-clinical decision roles, and privacy remains a recurring concern [13].

The present study examined a Temi V3 robot in the second-floor outpatient area of Tallinn Children’s Hospital in Tallinn, Estonia. The Temi V3 is a mobile platform with screen-based interaction, speech output, and navigation capabilities [14]. The robot was introduced as a public-facing assistive technology for wayfinding, simple informational support, and child-friendly engagement during waiting. Related earlier work by the research team examined scenarios for telepresence robots in health care [15]; the present manuscript reports a separate 2025 field pilot focused on a SAR in a pediatric outpatient setting.

Drawing these strands together, the study used 4 sensitizing domains: role clarity, interaction affordance, social mediation, and organizational embedding. Role clarity concerns whether the robot has an obvious and locally meaningful purpose. Interaction affordance concerns whether visitors and staff can understand how to start, continue, benefit from, and recover from interaction. Social mediation concerns whether signage, demonstration, and visible endorsement reduce uncertainty and invite use. Organizational embedding concerns workflow, training, support, maintenance, content management, and responsibility. The 4 domains were developed as an author synthesis of TAM, UTAUT, the Almere model, and implementation evidence [10,11,16-20]. They guided deductive coding and mixed methods interpretation, while inductive analysis remained open to recurrent issues raised by staff. The framework is provisional and is not presented as a validated scale.

The aim of the study was to examine how contextual support shaped observable visitor interaction with Temi V3 and which sociotechnical conditions staff associated with its practical acceptance in pediatric outpatient care. The following research questions (RQs) guided the study:

The study used a mixed methods pilot design with separate quantitative and qualitative analyses and convergent interpretation [21]. The quantitative strand examined spontaneous visitor interaction with the robot under 3 sequential implementation conditions. The qualitative strand explored how staff perceived the robot’s usefulness, limits, and implementation requirements through small focus group interviews. The strands were analyzed separately and then interpreted together to relate observable visitor behavior to staff perspectives on implementation, without treating staff accounts as evidence of visitors’ own perceptions.

This design was selected because unobtrusive observation was needed to document what people actually did near the robot under routine hospital conditions, while interviews were needed to understand how staff interpreted the robot’s role, usefulness, support burden, and fit with everyday work. The study did not link the 2 datasets at the person level. It used them as complementary evidence about immediate visitor uptake and organizational implementation.

The participant groups were intentionally different. The observation strand included naturalistic visitor passages, whereas the qualitative strand included employees with direct experience of using, observing, supporting, or coordinating the robot. Children, parents or guardians, and other adult visitors were not interviewed; claims about these groups are therefore limited to observed behavior and staff perceptions. RQ1 was addressed through condition-level behavior counts, observer-assigned age categories, and contemporaneous field notes, and RQ2 through staff interviews. Integration occurred after the separate analyses.

The study took place in the second-floor outpatient lobby at Tallinn Children’s Hospital. Families in this area navigate several specialties and may encounter uncertainty about room locations, changing numbering systems, and waiting routines. The robot’s intended public-facing functions included wayfinding, basic hospital information, and child-friendly engagement during waiting.

The robot’s pilot interface combined hospital-specific and generic elements in the national language (Estonian). Mapped hospital destinations and selected hospital information were available alongside generic Temi content, including news and simple entertainment. The robot’s hardware, mapped destinations, screen content, and physical location remained unchanged across the 3 observation conditions. The experiment’s conditions differed only in the external contextual support provided.

The study comprised anonymous, noninvasive observation and voluntary staff interviews. Ethics approval for observation was granted by the Tallinn Children’s Hospital Research Ethics Committee (TCH/2025-01), and the interview component was reviewed by Tallinn University of Technology’s Academic Ethics Research Committee. Because observation involved no recordings or identifiable personal data, individual consent was waived for the public space observation. Staff participants provided informed consent. Interview recordings and transcripts were stored securely, and the manuscript reports only aggregated observations and anonymized themes. All procedures complied with the General Data Protection Regulation (European Union 2016/679).

Across the 3 observation phases, 675 visitor passages were recorded, comprising 398 child and 277 adult passages. Each of the 4 separately recorded behavior indicators increased from baseline to poster support and again to staff facilitation (Table 1). The sums of the 4 indicator counts were 14, 52, and 156, respectively, corresponding to 5.4, 32.5, and 60.5 recorded indicators per 100 passages. These descriptive summaries include overlapping behaviors and are not percentages of passages with any interaction.

In the baseline phase (n=257), the robot was largely ignored. Occasional automatic voice output did not translate into sustained use. Looking was recorded for 8 (3.1%) passages, while approaching, touching, and screen or app use were each recorded for 2 (0.8%) passages.

When the poster was added, visible interaction increased (n=160). Looking was recorded for 30 (18.8%) passages, approaching for 12 (7.5%) passages, touching for 6 (3.8%) passages, and screen or app use for 4 (2.5%) passages. The sign provided the robot’s name, purpose, and interaction cues without changing its technical capabilities; whether visitors found these cues clearer or more useful was not assessed.

The highest counts occurred when a staff mediator stood near the robot and explicitly invited use (n=258). Looking was recorded for 80 (31.0%) passages, approaching for 44 (17.1%) passages, touching for 18 (7.0%) passages, and screen or app use for 14 (5.4%) passages. In descriptive terms, the phases represented progressively stronger levels of contextual and social support.

Contemporaneous field notes indicated that children were the most visibly responsive users in each phase, initiating contact through looking, approaching, touching, or play, while adults more often engaged in a supervisory or instrumental manner, for example, by watching a child interact or asking what the robot was for. Because age category was not linked to person-level behavior codes, these observations do not establish age-specific engagement rates and are not presented as a statistical comparison.

Observed indicator counts were higher with poster support and highest with staff facilitation. Staff accounts were consistent with the potential importance of clear functions and human prompting but did not establish why visitors interacted or whether they experienced less uncertainty. The strands provide complementary evidence about observable behavior and staff perceptions of implementation, rather than direct confirmation of visitors’ acceptance.

Field notes described children’s visible responsiveness, while staff accounts identified perceived benefits and organizational needs. Together, these findings informed the provisional mediated acceptance framework in Table 2. The framework organizes possible links between visitor-facing support and staff-identified implementation conditions; it has not been validated against direct visitor feedback or sustained service outcomes.

This mixed methods pilot examined how a socially assistive robot was received in a pediatric outpatient setting and which conditions shaped practical acceptance. Across all 4 observed indicators (looking at the robot, approaching it, touching it, and using the robot’s screen or app), interaction was lowest when the robot was present without explanation, increased when a poster clarified its role, and was highest when an intern invited and demonstrated use. Staff accounts provided a complementary implementation perspective rather than a direct explanation of visitors’ experiences. Staff evaluated the robot as a service tool whose value depended on a concrete role, relevant and multilingual content, technical support, workflow fit, and visible organizational ownership.

Taken together, observed visitor behavior and staff perceptions informed the provisional mediated acceptance framework. The robot’s hardware, location, and interface did not change across the 3 field phases; the surrounding support did. The results should nevertheless be read as descriptive associations. The sequential naturalistic design does not establish that the poster or facilitator alone caused the observed differences.

The stepwise increase in recorded behavior can be interpreted in light of technology acceptance theory. Clearer information and human facilitation are possible mechanisms consistent with perceived usefulness, ease of use, social influence, and facilitating conditions [16,17]. These constructs were not measured among visitors, so this interpretation remains theoretical rather than an empirical explanation of their behavior.

The stronger visible responsiveness of children also aligns with pediatric and hospital robotics literature. Earlier reviews suggest that children often engage readily with social robots when interaction is concrete, emotionally legible, and easy to interpret [3,8]. Staff interviewees described movement, voice, games, and distraction during waiting as meaningful for children. Staff evaluated the robot in terms of usefulness, reliability, and support burden, while field notes described adult visitors more often taking supervisory or instrumental roles. These observations and staff accounts do not establish differences in perceptions between children and adult visitors, although they suggest questions relevant to research on the broader care ecosystem [7,9]. The present age-related interpretation remains cautious because it is based on field notes and lacks linked age-by-behavior counts.

The findings also support a sociotechnical interpretation of acceptance. Staff did not ask for a more spectacular or more human-like robot. They asked for clearer functions, stronger content, multilingual capacity, easier troubleshooting, training, and responsibility. These priorities mirror barriers and enablers identified in implementation reviews, where maintenance, workflow fit, reliability, and stewardship influence whether robots move beyond pilot visibility [10,11,19,20]. Staff descriptions of child engagement and usability constraints raise hypotheses relevant to enjoyment, social presence, trust, and ease of use in the Almere model [18]. These constructs were not assessed directly among visitors and should not be treated as observed acceptance outcomes.

The integrated framework adds a practical bridge between these traditions (Table 2). TAMs explain why a person may approach and try the robot, while implementation research explains whether the organization can sustain the service. In this study, observation documented behavior under different support conditions, while staff accounts identified perceived longer-term needs such as content management, troubleshooting, and ownership. The contribution is a provisional synthesis of these complementary perspectives, not evidence of visitors’ perceived usefulness, trust, or acceptance.

The descriptive observation patterns and staff accounts suggest that hospitals could evaluate clear signage, explicit interaction cues, and early human facilitation when introducing SARs. These approaches may support visitor interaction, but their effects on visitors’ understanding, uncertainty, and perceived usefulness were not assessed in this study. Staff-identified wayfinding, waiting time engagement, parent guidance, and simple procedural information as potential use cases. These suggestions are consistent with roles discussed in the broader health care robot literature [4,5], but the present observations did not establish successful task completion or benefits perceived by visitors. Such functions should therefore be evaluated with children, parents or guardians, and other adult visitors before broader implementation.

The interviews also suggest that a child-facing function may create indirect value for other people in the care environment. A robot that holds a child’s attention during waiting may make navigation or preparation easier for parents and may reduce some routine interaction demands on staff. Future deployments should assess this potential chain of benefit directly; visible engagement alone is insufficient. Hospitals should therefore define expected outcomes for each use case, such as successful wayfinding, reduced uncertainty, or perceived helpfulness, and select measures that correspond to those outcomes. Staff accounts emphasized the need for organizational anchoring. A hospital should, with the help of relevant stakeholders, define what the robot does, where its task begins and ends, who updates content, who responds to faults, and how value will be assessed. Staff onboarding should include routine use and common recovery situations. The framework in Table 2 summarizes staff-informed priorities for implementation; whether these priorities improve visitor experience or sustained use requires further evaluation.

This was a single-site pilot, and the quantitative strand used descriptive observation without inferential testing. The 3 conditions were sequential and conducted on different dates, so temporal variation in visitor flow or other unmeasured contextual factors may have contributed to the differences. Counts represented anonymous passages, and repeat passages could not be identified. The 4 behavior indicators could overlap; their summed count per 100 passages does not measure the proportion of visitors who interacted, the depth or duration of engagement, or acceptance. Age category was not linked to individual behavior codes; therefore, age-related observations were limited to descriptive interpretation of the field notes.

The qualitative strand involved a small, staff-focused sample that differed from the observed visitor population. Children, parents or guardians, and other adult visitors were not interviewed. Although staff had direct exposure to the robot, their accounts reflect their own perceptions and cannot establish visitors’ perceived usefulness, ease of use, reasons for interacting, or acceptance. Because the datasets were not linked at the person level, integration cannot establish that staff-identified factors explain the observed behavioral differences. Exact staff age, gender, tenure, and job title distributions were not collected to protect the identity of employees. The study also did not directly measure child emotion, anxiety, trust, or satisfaction, although affect is central to social HRI [8,26]. These limitations restrict transferability and indicate that the 4-domain framework remains provisional.

Future research should combine in situ observation with ethically appropriate direct feedback from children, parents, adult visitors, and staff. Repeated or counterbalanced implementation periods and privacy-preserving person-level behavior coding would permit stronger comparison of contextual and age-related patterns. Longer follow-up should examine whether multilingual content, stronger interaction design, staff onboarding, explicit ownership, and maintenance arrangements support sustained use after novelty effects diminish.

Comparative work should also examine different robot-mediated tasks and settings. Wayfinding may require clear movement, stopping, and destination cues, while waiting time engagement may require age-appropriate content, audibility, and careful limits on distraction. Multisite studies could distinguish context-specific findings from more general implementation requirements and test whether the 4-domain framework remains useful across pediatric hospitals with different layouts, languages, staffing models, and patient populations.

In this single-site descriptive pilot, observed visitor interaction with Temi V3 was higher under poster support and staff facilitation than at baseline. Staff perceived potential value in wayfinding, child engagement during waiting, and basic guidance, and they identified content, language coverage, reliability, support, and workflow fit as important implementation considerations.

These complementary findings informed a provisional mediated acceptance framework comprising role clarity, interaction affordance, social mediation, and organizational embedding. The framework is an interpretation of observed behavior and staff perspectives, not a demonstration of visitors’ acceptance or of the effects of specific support mechanisms. Direct feedback from children, parents or guardians, and other adult visitors, together with longer-term evaluation, is needed to assess these proposed conditions and their contribution to sustained use.